Has Xbitlabs already benched Bulldozer?

[Neutralman]

I love reading your posts Looncraz!

 

[BlueBlazer]

So what if Lenovo is a chinese corporation? Their thinkpad T and X lines are seriously top of the line in business grade laptop quality, they have the same hardware (intel made chips, matte screen, etc..) as other big brands business line, yet lenovo manages to deliver it at a MUCH better price point.

When Thinkpads were IBM machines, the prices are much higher (literally premium class prices). This brings me to the next point below....

And last I checked, my family has 2 Dell latitude(business grade) laptops, 1 apple macbook, 1 sony vaoi laptop, 1 toshiba, and my own thinkpad. THEY ARE ALL MADE IN CHINA

Dell, Sony, Apple and Toshiba are well known brand names compared to Lenovo. They are all made in China, and what differentiates them is "branding". That alone can dictate the price of the laptops, and Apple being now the leading brandname enjoys that status. If you want an analogy, then just travel to China and check out the handphone store. You will find handphones (with never-seen-before brands) that looks like and functions just like any Nokia or Sony, but yet way much cheaper (than the Nokia and Sony handphones of the same calibre in the same store). Some look so much alike that you could have mistaken them for Nokia or Sony. They are all made in China.

You don't need to degrade us just because we are new posters of Anandtech. Looncraz presented some very well-thought arguments and analogies. Lenovo may make a lot less net profit per thinkpad sold, but their sales volume and quality of product make them very profitable.

I don't think so, even after all those typing. Laptops are manufactured by OEMs (e.g. Compal) and its one time cost. This is not the same as CPU market, as their cost and operating structures are different (as Arkaign points out). A lot more things have to be taken into account in CPU business (such as R&D, royalties, wafer costs, ramp cycles, yields, etc). :hmm:

So you must be the kind of person who thinks everything that is made in China is a cheap POS?

Even the iPad is made in China. Look at my comments above.

AMD has done this before, and the guy now running the show made his career on doing exactly this. If you are getting an extra $80/cpu for your launch, winning mind-share and reviews, the demand will go up and the price per lot will increase......

I think Arkaign answered that again. Might I add that you need to check AMD's ASPs. You will find that a large chunk of current desktop sales belongs to Athlon IIs. That should give you a general idea. Many (like yourself) may "appreciate" AMD because of their pricing ("bang-per-buck") and yet not realize that AMD is literally stuck at the bottom half tier. :hmm:

Well, I think you're over-thinking things somewhat, but I can say with all seriousness that you're an original, entertaining character

I agree! Guess everyone have their own opinions, idiosyncrasies and dreams.

 

[GaiaHunter]

I love how people conveniently forget that most desktop workloads use two-four threads.

What is the size of those extra 4 cores? How big is BD compared to SB?

 

[beginner99]

A SB core is much bigger than a BD core so that also tells you why single-threaded performance will certainly be lower than with SB.

Anyone read the Ivy Bridge Post? I found it interesting that one of the changes made was to speed up FP division considering AMD is focussing on Integer performance....

 

[Silverforce11]

A large chunk of SB and most likely IvB is dedicated to the GPU, so that will be a factor in terms of perf/mm.

 

[grkM3]

If bd is a monster I hope the new ceo prices them 700-900 so AMD can start making some money again.

Im sick of AMD being the cheaper alternative to intel and miss the 1000 opteron fx days where AMD was on top.

We have been waiting a long time for bd and it looks like its another mess from amd like everything else Iv been waiting for.

 

[BlueBlazer]

If bd is a monster I hope the new ceo prices them 700-900 so AMD can start making some money again.

Im sick of AMD being the cheaper alternative to intel and miss the 1000 opteron fx days where AMD was on top.

The correct phrase is "hope AMD can have the same pricing parity with Intel". Not necessarily talking about $1000 CPUs. Right now most of AMD's desktop CPUs are under $200.

 

[sequoia464]

What would the differences be on the 6100 compared to a 8120. I know there are no benchmarks yet, but that hasn't stopped any educated conjecture so far...

If someone were to foolishly pre-order (not me of course) would they be making a mistake with a 6100, or would it possibly have the potential to clock higher than an 8120 because of the fewer cores? - would the loss of cores solidly negate the hypothetical extra MHZ??

I don't personally have a clue here, just looking for a some educated input. Thanks.

 

[GTRagnarok]

The correct phrase is "hope AMD can have the same pricing parity with Intel". Not necessarily talking about $1000 CPUs. Right now most of AMD's desktop CPUs are under $200.

ALL their desktop CPUs are under $200. It's a sad state :\

 

[ransomlist]

What would the differences be on the 6100 compared to a 8120. I know there are no benchmarks yet, but that hasn't stopped any educated conjecture so far...

I'm curious about this too.

The pessimist in me says 'its a tri core pretending to be a hex and it will get stomped by the i5-2500k and not exceed Thuban performance at stock'

The optimist says 'That thing will OC like a monster at a reasonable price and single thread perf shouldn't be much different to 4 module FX-8'

At the prices listed for pre-order, after 20% VAT in the UK, if an i5 gives significantly better perf i cant see anyone buying into FX-6x's.

 

[frozentundra123456]

Well, I think you're over-thinking things somewhat, but I can say with all seriousness that you're an original, entertaining character

Welcome to AT!

In addition, Intel has a huge reserve of cash. AMD is barely surviving. There is no way that Intel could not demolish AMD in a price war if AMD decided to start one. No matter how much AMD cut their prices, Intel could at least temporarily charge less for a similar preformance product.


In fact, I think Intel could easily drive AMD out of business is they desired, but are afraid of anti-trust intervention.

 

[GaiaHunter]

In addition, Intel has a huge reserve of cash. AMD is barely surviving. There is no way that Intel could not demolish AMD in a price war if AMD decided to start one. No matter how much AMD cut their prices, Intel could at least temporarily charge less for a similar preformance product.

Intel doesn't want a price war - it is bad for the price of Intel stock.

 

[looncraz]

What is the size of those extra 4 cores? How big is BD compared to SB?

AMD said the extra integer core added 12% to the size of the module. So not terribly much.

Bulldozer is smaller than the Phenom II X6 CPUs by just a touch.

What would the differences be on the 6100 compared to a 8120. I know there are no benchmarks yet, but that hasn't stopped any educated conjecture so far...

Similar to X4 vs X3. The additional L3 available per module will provide some benefit in streaming computations (low-threaded), but the lack of those extra cores will cost you in heavily threaded computations.

However, with Bulldozer, I'd subtract one core due to internal module overhead (10% per core, reducing frequency) and pretend you are getting a 5-core CPU with an extra thread... That should keep performance expectations under control.

AMD's Turbo 2.0, supposedly, can be set to a certain TDP target. So if you were to set a 6-core to 125W, you would get free, dynamic, overclocking - and likely remain boosted even under heavy loads. Sadly, however, I feel we will be limited to the stated max TDP of the CPU, with any official utility

Lenovo is formerly IBM's PC and laptop division sold off to a Chinese corporation (Lenovo is part of a multinational holdings/group from China). Their products can be as cheap as you can get.

As has already been stated - EVERYONE uses Chinese manufacturing.

There is an upside, though!! If you know what you're looking for, you can find some surprising finds on eBay. For instance, I own a $1200 high-end Pioneer 7" touch-screen double din car radio with 3D GPS, Bluetooth, backup camera, 3D menus, WinCE 5.0, TV receiver, and MUCH more for a mere ~$250.

Yes, it is the SAME unit Pioneer sales for $1200! The software is *slightly* different, and I have EXTRA hardware features... but it is the same (except the place where Pioneer glues in their logo is empty ;-) ). It even comes on and says "Pioneer," but I changed it and now it says Volvo

It is a quality unit, for what I feel is - finally - a fair price.

Besides, what you're missing is that if you are already going to be making 50% more per die than before, underselling the first thousand chips so the next ten thousand sell at higher premiums due to "inflated demand" isn't very difficult to envision or accomplish.

I feel it is no coincidence that the rumored prices fell by $40/CPU with the new CEO claiming a new AMD strategy of "being the predator." Predators are cunning, they will do their best to get more from less. Besides, these people aren't exactly high school students...some are even as smart as me!! :whiste:

--The loon

 

[sandorski]

Re Pricing and Performance: If BD is superior to what's available Now, but AMD knows that Intel will counter and take back the top spot in a short period of time afterwards, then it makes sense for the Pricing to be reasonably forward looking. Otherwise people will just wait for Intel's soon to be released Product. AMD needs to move Product in Volume to succeed.

 

[MODist]

Intel doesn't want a price war - it is bad for the price of Intel stock.

Record profits are bad for Intel stock. Doesn't matter what they do their stock goes nowhere.

 

[Fox5]

So from what I could find by Googling, a Bulldozer core (not module) is about the same size as a Phenom II core, and still significantly smaller than Sandy Bridge. In that case, I'd hope IPC is up, and multithreaded performance is significantly up.

 

[JFAMD]

Spot on, but how much wiggle room is there for a website to make a statement like the one quoted in the original post?

Technically they haven't released any information, but they've implied the BD is going to be more of a monster than a lot of people have been speculating. I don't know what kind of relationship Xbitlabs has with AMD or if they're a likely site to receive hardware to benchmark and review, but it's either the case that they have received and benchmarked a production quality chip or that they've made a clever quip designed to drive page views and fan the flames of speculation.

How much wiggle room?

You'll know when you don't get a chip next time.

I would assume it is not worth the risk.

All of the press guys at AMD for the overclocking event had NDAs that said you cannot say anything until the press release went live.

How many of them posted anything prior?

I saw one story that was from a site that was not invited.

The long term gain of keeping an NDA is far more valuable than blurting out a comment ahead of time. There is little or no short-term gain, and plenty of downside.

 

[LOL_Wut_Axel]

So from what I could find by Googling, a Bulldozer core (not module) is about the same size as a Phenom II core, and still significantly smaller than Sandy Bridge. In that case, I'd hope IPC is up, and multithreaded performance is significantly up.

Even if IPC is up, we have to take into account the loss from the resource sharing, especially in floating point-heavy programs like POV-Ray. In gaming and audio encoding it'll obviously be slower than Sandy Bridge unless they can get 1GHz or more higher OCs or IPC within 15% of Sandy Bridge, which I doubt.

Personally, the multi-threaded programs I care about it performing the best it can is Visual Studio, x264 HD Pass Two, and Handbrake. These are also programs the Phenom II X6 did great at, so we'll see.

 

[JFAMD]

If you think that there is somehow a problem in FP-heavy applications, I'd love to understand how that is.

Each module has a dedicated FP scheduler. The intel architecture does FP scheduling from a shared integer/FP scheduler. The integer thread, the hyperthread and the FP execution all share a 56-entry scheduler.

We have a 40-entry scheduler for each integer core and a dedicated 60-entry scheduler for the FP pipe.

So, if you want to talk about the impact on FP performance from shared resources, you're barking up the wrong tree.

 

[Mopetar]

How much wiggle room?

You'll know when you don't get a chip next time.

I would assume it is not worth the risk.

All of the press guys at AMD for the overclocking event had NDAs that said you cannot say anything until the press release went live.

How many of them posted anything prior?

I saw one story that was from a site that was not invited.

The long term gain of keeping an NDA is far more valuable than blurting out a comment ahead of time. There is little or no short-term gain, and plenty of downside.

You're pretty good at what you do.

That said, hopefully we'll get some real answers to all of our performance questions sooner rather than later. While some of us like to shake our presents before opening them, eventually we'd like Christmas to actually arrive.

 

[LOL_Wut_Axel]

If you think that there is somehow a problem in FP-heavy applications, I'd love to understand how that is.

Each module has a dedicated FP scheduler. The intel architecture does FP scheduling from a shared integer/FP scheduler. The integer thread, the hyperthread and the FP execution all share a 56-entry scheduler.

We have a 40-entry scheduler for each integer core and a dedicated 60-entry scheduler for the FP pipe.

So, if you want to talk about the impact on FP performance from shared resources, you're barking up the wrong tree.

You were the ones stressing that FP performance wasn't as important in desktop workloads, and therefore that's where most of the compromise when sharing resources comes from.

The two integer units in a module are equivalent to two normal ones, but the change comes when you look at the floating point units. If there's a request to handle FP execution on only one of the two cores in amodule, it has the 2x 128-bit FMAC units to itself and can handle 256-bit AVX instructions normally. If there's a request to handle FP execution on both cores, then obviously some performance will be lost. How much exactly, you guys haven't told us.

 

[NostaSeronx]

You were the ones stressing that FP performance wasn't as important in desktop workloads, and therefore that's where most of the compromise when sharing resources comes from.

The two integer units in a module are equivalent to two normal ones, but the change comes when you look at the floating point units. If there's a request to handle FP execution on only one of the two cores in a module, it has the 2x 128-bit FMAC units to itself and can handle 256-bit AVX instructions normally. If there's a request to handle FP execution on both cores, then obviously some performance will be lost. How much exactly, you guys haven't told us.

The purpose of sharing is hiding latency not introducing latency

Having shared components allows for queuing

Do you think Phenom II can actually feed its Floating Point Coprocessor if it was upgraded to "256bit" AVX? Without introducing latency? Forbid if it was a benchmark that only accepted "256bit" AVX

By having the Floating Point Coprocessor shared there will always be something that the Floating Point Unit can do

AMD has told us a lot of things in their software guide

Only 1 256-bit operation can issue per cycle, however an extra cycle can be incurred as in the case
of a FastPath Double if both micro ops cannot issue together.

I do think this is talking about Macro ops

Because

FastPath Double Decodes directly into two macro-ops in microprocessor hardware.

A macro-op and a micro-op can either be simple(32bit) or complex(64bit)

A macro op includes 1 Int or 1 Floating Point and 1 Load or 1 Store or 1 Load and Store to the Same Address

Then another thing:

The AMD Family 15h processor floating point unit (FPU) was designed to provide four times the raw
FADD and FMUL bandwidth as the original AMD Opteron and Athlon 64 processors.

So, in Floating Point it has 4x the raw bandwidth over K8 FXs

I hope I answered some peoples questions...

 

[Abwx]

You were the ones stressing that FP performance wasn't as important in desktop workloads, and therefore that's where most of the compromise when sharing resources comes from.

The two integer units in a module are equivalent to two normal ones, but the change comes when you look at the floating point units. If there's a request to handle FP execution on only one of the two cores in amodule, it has the 2x 128-bit FMAC units to itself and can handle 256-bit AVX instructions normally. If there's a request to handle FP execution on both cores, then obviously some performance will be lost. How much exactly, you guys haven't told us.

As said ad nauseam , there s four 128b FP pipe lines...:biggrin:

 

[JFAMD]

The two integer units in a module are equivalent to two normal ones, but the change comes when you look at the floating point units. If there's a request to handle FP execution on only one of the two cores in amodule, it has the 2x 128-bit FMAC units to itself and can handle 256-bit AVX instructions normally. If there's a request to handle FP execution on both cores, then obviously some performance will be lost. How much exactly, you guys haven't told us.

But you keep going back to AVX-256. Are we talking about client or server. I don't expect to see that anywhere on the client side and only rarely on the server side.

You are aware that FP pipelines are really long, right? And that if you mix AVX and SSE, intel needs to clear the pipeline completely from one before issuing the other, right?

With flex FP you can run one AVX and one SSE at the same time.

Check out last year's IDF presentations, there was one that specifically talked about recoding all of your SSE to AVX in order to prevent this from happening. If client applications are not filling the FP pipes today, I see little opportunity for anyone to change their code to reflect AVX; most of the client apps will probably stay with SSE for now.

I am just trying to figure out where exactly (on which workloads) you believe this is going to be an issue.

I contend on client workloads that SSE and AVX will be immaterial, but on server workloads there could be a difference (and FMA4/XOP will help us.) I am not interested in a discussion of client FP benchmarks, I'd like to discuss real apps.

 

[LOL_Wut_Axel]

But you keep going back to AVX-256. Are we talking about client or server. I don't expect to see that anywhere on the client side and only rarely on the server side.

You are aware that FP pipelines are really long, right? And that if you mix AVX and SSE, intel needs to clear the pipeline completely from one before issuing the other, right?

With flex FP you can run one AVX and one SSE at the same time.

Check out last year's IDF presentations, there was one that specifically talked about recoding all of your SSE to AVX in order to prevent this from happening. If client applications are not filling the FP pipes today, I see little opportunity for anyone to change their code to reflect AVX; most of the client apps will probably stay with SSE for now.

I am just trying to figure out where exactly (on which workloads) you believe this is going to be an issue.

I contend on client workloads that SSE and AVX will be immaterial, but on server workloads there could be a difference (and FMA4/XOP will help us.) I am not interested in a discussion of client FP benchmarks, I'd like to discuss real apps.

Client.

But thanks for the explanation, helps clear some things up. I guess we'll finally see in a month if there's a performance deficit or not.